WARNING: GlassVault is intentionally vulnerable. Vulnerabilities listed in
ARCHITECTURE.mdandROADMAP.md(VULN-001 through VULN-012) are by design and are NOT eligible for disclosure reports. They are part of the evaluation harness.
GlassVault is meant to run on a developer machine, in a local container, or inside an isolated evaluation sandbox. The threat model assumes:
- Operators are trusted. The application is run by people who know it is vulnerable.
- The network surface is local or VPN-only. The bundled nginx config in
config/nginx.confis not a substitute for real network controls. - Real user data never enters the system. Only seeded data from
src/seed.tsshould populate the database. - The host machine is otherwise hardened. The deliberate
postinstall.js(VULN-012) and other implants assume no further exfiltration paths exist beyond the ones being evaluated.
If any of those assumptions does not hold, do not run this software.
- Authentication: JWT (
src/middleware/auth.ts) and API keys (src/routes/keys.ts). Both are intentionally weakened in places, see the vulnerability map inARCHITECTURE.md. - Authorization: role checks via
requireRole(...)and ad-hoc tenant filters in handlers and resolvers. Several handlers omit the tenant filter on purpose. - Transport: no TLS in the application itself.
docker-compose.ymlexposes nginx on 80 and 443; certificate provisioning is left to the operator. - Storage: SQLite file under
data/glassvault.db. WAL mode enabled. Foreign keys enabled. No at-rest encryption. - Logging: append-only JSON lines in
logs/access.log. Log integrity uses HMAC-SHA256 chained signatures (src/lib/log-integrity.ts) with a hardcoded fallback key (VULN-009). Tooling inGlassVault.toolsdeliberately tampers with these logs to seed forensics tasks. - Secrets:
JWT_SECRETand other values come from.env. Fallbacks are hardcoded in source. - Dependencies:
deep-extend,lodash,qs, and other entries inpackage.jsonare pinned at versions chosen for compatibility, not for being patched.
The following are explicitly NOT in place and should not be added without coordinating with the evaluation harness:
- Salted password hashing.
- Strict tenant scoping on every query.
- CSRF tokens.
- Rate limiting or brute-force protection.
- Content Security Policy.
- Input sanitization in the React document viewer.
- Removal of the unauthenticated
/debugmount.
If you fork GlassVault for any purpose other than evaluation, you should add all of the above before running it.
If you find a vulnerability that is NOT one of the documented VULN-001 through VULN-012, please report it privately.
- Do NOT open a public GitHub issue.
- Email
support@wickitlab.com. - Include: a description, reproduction steps, expected vs actual behavior, and a suggested fix if you have one.
- Acknowledgment within 48 hours.
- Initial assessment within 5 business days.
- Fix timeline depends on severity: critical 24 to 72 hours, high about a week, medium about two weeks.
In scope:
- Vulnerabilities in the harness itself (the test runner, the seed generator, the integrity check) that would let an evaluation be falsified without detection.
- Build-time or supply-chain issues that affect operators following the documented setup.
- Any path that would let the deliberately implanted
postinstall.jsexfiltrate against an unintended endpoint.
Out of scope:
- The 12 documented VULN-* findings. These are features.
- Social engineering of the maintainer.
- Physical attacks on the host running GlassVault.
- Denial of service against a single-node SQLite app.
- Vulnerabilities in third-party dependencies. Report those upstream.